A protein’s shape helps determine what it can do. Predicting that shape from its amino-acid sequence was an open problem of biology.
At CASP14, a rigorous blind assessment, AlphaFold achieved accuracy competitive with experimental structures for many targets. Its neural network considers evolutionary, physical and geometric signals, then directly predicts atomic coordinates.
The achievement does not eliminate laboratory work: proteins move, interact and respond to their environment, but it changes the starting point. Researchers can then approach a protein with a high-quality structural hypothesis in hand.
Structure can reveal how proteins work, where drugs might bind and how a mutation may alter its function. This compresses the early stages of investigation.
Highly accurate protein structure prediction with AlphaFold, 596, 583–589
Nature · 2021
Read the primary source ↗